AI and the Labour Market in Croatian Media

Media Framing Analysis (2021–2023)

Abstract

This document analyses how Croatian digital media frame the intersection of artificial intelligence and the labour market. The corpus is drawn from the Determ media-monitoring platform (January 2021 through December 2023) and includes only articles containing both AI-related and labour-market-related keywords (intersection logic). Eight interpretive frames, six actor categories, and automated sentiment are examined across time, platform, and media outlet types. The analysis provides a comprehensive exploratory and descriptive foundation for subsequent econometric work.

NoteData preparation

The corpus was extracted by R/01_extract_corpus.R and enriched by R/02_add_diagnostics.R. Run those scripts first if the data files are missing.

1 Introduction

1.1 Background

Technological change has historically reshaped labour demand in ways that were difficult to predict at the time. Artificial intelligence poses a distinctive challenge because it affects not only routine manual tasks but also non-routine cognitive work such as programming, translation, data analysis, and creative writing.

Media do not merely relay facts; they select, omit, and emphasise, thereby actively shaping public perception. The way media frame the impact of AI on work can influence individual behaviour, corporate strategy, and public policy.

1.2 Research questions

RQ1 (Volume and dynamics) How much media coverage does the AI–labour nexus receive, and how has coverage evolved over time?

RQ2 (Frames) Which interpretive frames dominate, and how has their prevalence shifted?

RQ3 (Actors) Whose perspectives appear most frequently?

RQ4 (Sources) Do different media outlet types frame the topic differently?

2 Data

2.1 Loading the corpus

if (!file.exists(path_raw_corpus)) {
  stop("Corpus not found at: ", path_raw_corpus,
       "\n\nRun R/01_extract_corpus.R first.")
}

corpus_data <- readRDS(path_raw_corpus)

# --- DATE FILTER: keep only articles before 2024-01-01 ---
corpus_data <- corpus_data |>
  filter(DATE < as.Date("2024-01-01"))

cat("Corpus loaded successfully\n")

Corpus loaded successfully

cat("Total articles:", format(nrow(corpus_data), big.mark = ","), "\n")

Total articles: 33,692 

cat("Date range:", as.character(min(corpus_data$DATE)), "to",
    as.character(max(corpus_data$DATE)), "\n")

Date range: 2021-01-01 to 2023-12-31 

cat("Columns:", ncol(corpus_data), "\n")

Columns: 14 

corpus_data$.text_lower <- stri_trans_tolower(
  paste(coalesce(corpus_data$TITLE, ""),
        coalesce(corpus_data$FULL_TEXT, ""),
        sep = " ")
)

if (!"year" %in% names(corpus_data)) {
  corpus_data$year <- year(corpus_data$DATE)
}
if (!"year_month" %in% names(corpus_data)) {
  corpus_data$year_month <- floor_date(corpus_data$DATE, "month")
}
if (!"quarter" %in% names(corpus_data)) {
  corpus_data$quarter <- quarter(corpus_data$DATE)
  corpus_data$year_quarter <- paste0(corpus_data$year, " Q", corpus_data$quarter)
}

corpus_data$word_count <- stri_count_regex(corpus_data$FULL_TEXT, "\\S+")

summary_stats <- tibble(
  Metric = c("Total articles", "Unique sources", "Date range",
             "Mean words/article", "Median words/article"),
  Value = c(
    format(nrow(corpus_data), big.mark = ","),
    format(n_distinct(corpus_data$FROM), big.mark = ","),
    paste(min(corpus_data$DATE), "to", max(corpus_data$DATE)),
    round(mean(corpus_data$word_count, na.rm = TRUE)),
    round(median(corpus_data$word_count, na.rm = TRUE))
  )
)

kable(summary_stats, col.names = c("Metric", "Value")) |>
  kable_styling(bootstrap_options = c("striped", "hover"), full_width = FALSE)

Metric	Value
Total articles	33,692
Unique sources	2,034
Date range	2021-01-01 to 2023-12-31
Mean words/article	899
Median words/article	646

2.2 Source distribution

if ("SOURCE_TYPE" %in% names(corpus_data)) {
  source_dist <- corpus_data |>
    count(SOURCE_TYPE, sort = TRUE) |>
    mutate(pct = round(n / sum(n) * 100, 1))

  kable(source_dist, col.names = c("Source type", "N", "%")) |>
    kable_styling(bootstrap_options = c("striped", "hover"), full_width = FALSE)
}

Source type	N	%
web	31927	94.8
facebook	727	2.2
youtube	433	1.3
forum	269	0.8
reddit	256	0.8
twitter	67	0.2
comment	13	0.0

top_sources <- corpus_data |>
  count(FROM, sort = TRUE) |>
  head(25)

ggplot(top_sources, aes(x = reorder(FROM, n), y = n)) +
  geom_col(fill = "#2c7bb6", alpha = 0.8) +
  coord_flip() +
  labs(title = "Top 25 sources", x = NULL, y = "Articles")

Figure 1: Top 25 sources by article count

2.3 Article length distribution

ggplot(corpus_data |> filter(word_count > 0, word_count < quantile(word_count, 0.99, na.rm = TRUE)),
       aes(x = word_count)) +
  geom_histogram(bins = 60, fill = "#2c7bb6", alpha = 0.7, color = "white") +
  geom_vline(xintercept = median(corpus_data$word_count, na.rm = TRUE),
             linetype = "dashed", color = "#d7191c") +
  annotate("label",
           x = median(corpus_data$word_count, na.rm = TRUE),
           y = Inf, vjust = 1.5,
           label = paste("Median:", round(median(corpus_data$word_count, na.rm = TRUE))),
           fill = "white", size = 3.5) +
  labs(title = "Article length distribution",
       subtitle = "Top 1% trimmed. Dashed line marks the median.",
       x = "Word count", y = "Articles")

Figure 2: Distribution of article length (word count)

if ("SOURCE_TYPE" %in% names(corpus_data)) {
  wc_by_source <- corpus_data |>
    filter(!is.na(SOURCE_TYPE)) |>
    group_by(SOURCE_TYPE) |>
    summarise(
      n       = n(),
      median  = round(median(word_count, na.rm = TRUE)),
      mean    = round(mean(word_count, na.rm = TRUE)),
      p25     = round(quantile(word_count, 0.25, na.rm = TRUE)),
      p75     = round(quantile(word_count, 0.75, na.rm = TRUE)),
      .groups = "drop"
    ) |>
    arrange(desc(median))

  kable(wc_by_source,
        col.names = c("Source type", "N", "Median", "Mean", "P25", "P75")) |>
    kable_styling(bootstrap_options = c("striped", "hover"), full_width = FALSE)
}

Table 1: Article length by source type

Source type	N	Median	Mean	P25	P75
web	31927	674	935	439	1105
forum	269	317	425	171	571
youtube	433	232	268	138	359
reddit	256	163	256	87	318
comment	13	95	150	83	149
facebook	727	92	149	55	158
twitter	67	36	35	32	39

2.4 Platform landscape

corpus_data <- corpus_data |>
  mutate(
    platform = case_when(
      tolower(SOURCE_TYPE) %in% c("web", "internet") ~ "web",
      tolower(SOURCE_TYPE) %in% c("facebook", "fb")  ~ "Facebook",
      tolower(SOURCE_TYPE) %in% c("twitter", "x")    ~ "Twitter",
      tolower(SOURCE_TYPE) == "instagram"             ~ "Instagram",
      tolower(SOURCE_TYPE) == "youtube"               ~ "YouTube",
      tolower(SOURCE_TYPE) == "tiktok"                ~ "TikTok",
      tolower(SOURCE_TYPE) == "reddit"                ~ "Reddit",
      tolower(SOURCE_TYPE) %in% c("forum", "forum.hr") ~ "forum",
      TRUE ~ "Other"
    )
  )

platforms_with_data <- corpus_data |>
  count(platform) |>
  filter(n >= 20) |>
  pull(platform)

platform_monthly <- corpus_data |>
  filter(platform %in% platforms_with_data) |>
  count(year_month, platform) |>
  filter(!is.na(year_month))

ggplot(platform_monthly, aes(x = year_month, y = n)) +
  geom_col(aes(fill = platform), alpha = 0.7, show.legend = FALSE) +
  geom_smooth(method = "loess", se = FALSE, color = "black",
              linewidth = 0.7, span = 0.3) +
  facet_wrap(~ platform, ncol = 2, scales = "free_y") +
  scale_fill_manual(values = platform_colors) +
  scale_x_date(date_breaks = "6 months", date_labels = "%b\n%Y") +
  labs(title = "Monthly article volume by platform",
       x = NULL, y = "Articles") +
  theme(axis.text.x = element_text(size = 8))

Figure 3: Monthly volume by platform

3 Temporal dynamics

3.1 Monthly volume

monthly_volume <- corpus_data |>
  count(year_month) |>
  filter(!is.na(year_month))

events_cfg <- CONFIG$events
events <- tibble(
  date  = as.Date(sapply(events_cfg, `[[`, "date")),
  label = sapply(events_cfg, `[[`, "label"),
  y_pos = max(monthly_volume$n, na.rm = TRUE) * seq(0.95, by = -0.10,
                                                     length.out = length(events_cfg))
)

# Keep only events within data range
events <- events |> filter(date <= max(corpus_data$DATE))

ggplot(monthly_volume, aes(x = year_month, y = n)) +
  geom_col(fill = "#2c7bb6", alpha = 0.7) +
  geom_smooth(method = "loess", se = TRUE, color = "#d7191c", linewidth = 1) +
  geom_vline(data = events, aes(xintercept = date),
             linetype = "dashed", color = "gray40") +
  geom_label(data = events, aes(x = date, y = y_pos, label = label),
             size = 3, fill = "white", alpha = 0.9) +
  scale_x_date(date_breaks = "3 months", date_labels = "%b\n%Y") +
  labs(title = "AI and labour market coverage",
       subtitle = "Monthly volume with LOESS trend and key events",
       x = NULL, y = "Articles")

Figure 4: Monthly article volume with key events

3.2 Yearly volume

yearly_volume <- corpus_data |>
  count(year) |>
  filter(!is.na(year))

ggplot(yearly_volume, aes(x = factor(year), y = n)) +
  geom_col(fill = "#2c7bb6", alpha = 0.8) +
  geom_text(aes(label = format(n, big.mark = ",")), vjust = -0.5) +
  labs(title = "Annual volume", x = "Year", y = "Articles") +
  expand_limits(y = max(yearly_volume$n) * 1.1)

Figure 5: Annual article volume

3.3 Quarterly breakdown

quarterly_volume <- corpus_data |>
  count(year, quarter) |>
  filter(!is.na(year)) |>
  mutate(yq = paste0(year, " Q", quarter)) |>
  arrange(year, quarter) |>
  mutate(
    lag_n = lag(n, 4),
    yoy_growth = ifelse(!is.na(lag_n) & lag_n > 0,
                        round((n / lag_n - 1) * 100, 1), NA_real_)
  )

p_qvol <- ggplot(quarterly_volume, aes(x = fct_inorder(yq), y = n)) +
  geom_col(fill = "#2c7bb6", alpha = 0.8) +
  geom_text(aes(label = n), vjust = -0.5, size = 3) +
  labs(title = "Quarterly volume", x = NULL, y = "Articles") +
  theme(axis.text.x = element_text(angle = 45, hjust = 1)) +
  expand_limits(y = max(quarterly_volume$n, na.rm = TRUE) * 1.1)

p_qgrowth <- quarterly_volume |>
  filter(!is.na(yoy_growth)) |>
  ggplot(aes(x = fct_inorder(yq), y = yoy_growth,
             fill = yoy_growth > 0)) +
  geom_col(alpha = 0.8, show.legend = FALSE) +
  geom_hline(yintercept = 0, color = "gray40") +
  geom_text(aes(label = paste0(yoy_growth, "%")),
            vjust = ifelse(quarterly_volume$yoy_growth[!is.na(quarterly_volume$yoy_growth)] > 0, -0.5, 1.5),
            size = 3) +
  scale_fill_manual(values = c("TRUE" = "#4daf4a", "FALSE" = "#e41a1c")) +
  labs(title = "Year-over-year growth rate", x = NULL, y = "% change") +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

p_qvol / p_qgrowth

Figure 6: Quarterly article volume with year-over-year growth

3.4 Daily volume heatmap

daily_volume <- corpus_data |>
  count(DATE) |>
  filter(!is.na(DATE)) |>
  mutate(
    wday = wday(DATE, label = TRUE, week_start = 1),
    week = floor_date(DATE, "week", week_start = 1)
  )

ggplot(daily_volume, aes(x = week, y = wday, fill = n)) +
  geom_tile(color = "white", linewidth = 0.3) +
  scale_fill_gradient(low = "#f7fbff", high = "#08306b",
                      name = "Articles") +
  scale_x_date(date_breaks = "3 months", date_labels = "%b\n%Y") +
  labs(title = "Daily article volume",
       subtitle = "Darker cells indicate higher activity",
       x = NULL, y = NULL) +
  theme(axis.text.y = element_text(size = 9))

Figure 7: Daily article volume (calendar heatmap)

3.5 Day-of-week patterns

weekday_avg <- corpus_data |>
  mutate(wday = wday(DATE, label = TRUE, week_start = 1)) |>
  count(DATE, wday) |>
  group_by(wday) |>
  summarise(avg = mean(n), .groups = "drop")

ggplot(weekday_avg, aes(x = wday, y = avg)) +
  geom_col(fill = "#2c7bb6", alpha = 0.8) +
  geom_text(aes(label = round(avg, 1)), vjust = -0.5, size = 3.5) +
  labs(title = "Publication pattern by day of week",
       x = NULL, y = "Average articles per day") +
  expand_limits(y = max(weekday_avg$avg) * 1.1)

Figure 8: Average daily articles by day of week

4 Frame analysis

4.1 Frame dictionaries

# Build dictionaries from config
frame_dictionaries <- lapply(CONFIG$frames, `[[`, "keywords")

frame_summary <- tibble(
  Frame       = names(frame_dictionaries),
  Description = sapply(CONFIG$frames, `[[`, "description"),
  Keywords    = sapply(frame_dictionaries, length)
)

kable(frame_summary, col.names = c("Frame", "Description", "N keywords")) |>
  kable_styling(bootstrap_options = c("striped", "hover"), full_width = FALSE)

Frame	Description	N keywords
JOB_LOSS	AI eliminates jobs	22
JOB_CREATION	AI creates new opportunities	13
TRANSFORMATION	AI transforms rather than destroys work	16
SKILLS	Focus on reskilling and education	14
REGULATION	Policy, governance, and worker protection	13
PRODUCTIVITY	Economic benefits and efficiency	11
INEQUALITY	Distributional concerns and digital divide	11
FEAR_RESISTANCE	Anxiety, fear, and opposition to AI	16

4.2 Frame detection

for (frame_name in names(frame_dictionaries)) {
  pattern <- paste(frame_dictionaries[[frame_name]], collapse = "|")
  corpus_data[[paste0("frame_", frame_name)]] <- stri_detect_regex(
    corpus_data$.text_lower, pattern
  )
}

frame_cols <- paste0("frame_", names(frame_dictionaries))
frame_counts <- corpus_data |>
  summarise(across(all_of(frame_cols), sum, na.rm = TRUE)) |>
  pivot_longer(everything(), names_to = "frame", values_to = "count") |>
  mutate(
    frame = str_remove(frame, "frame_"),
    pct   = round(count / nrow(corpus_data) * 100, 1)
  ) |>
  arrange(desc(count))

kable(frame_counts, col.names = c("Frame", "Articles", "% corpus")) |>
  kable_styling(bootstrap_options = c("striped", "hover"), full_width = FALSE)

Frame	Articles	% corpus
PRODUCTIVITY	7748	23.0
FEAR_RESISTANCE	5695	16.9
SKILLS	2372	7.0
TRANSFORMATION	1755	5.2
JOB_CREATION	1550	4.6
REGULATION	1519	4.5
INEQUALITY	664	2.0
JOB_LOSS	651	1.9

ggplot(frame_counts, aes(x = reorder(frame, count), y = count, fill = frame)) +
  geom_col(alpha = 0.8) +
  geom_text(aes(label = paste0(pct, "%")), hjust = -0.1, size = 3.5) +
  coord_flip() +
  scale_fill_manual(values = frame_colors, guide = "none") +
  labs(title = "Interpretive frame prevalence",
       subtitle = "Percentage of articles containing frame keywords",
       x = NULL, y = "Articles") +
  expand_limits(x = max(frame_counts$count) * 1.15)

Figure 9: Frame prevalence

4.3 Frame evolution

frame_monthly <- corpus_data |>
  group_by(year_month) |>
  summarise(
    n_total = n(),
    across(all_of(frame_cols), sum, na.rm = TRUE)
  ) |>
  pivot_longer(cols = all_of(frame_cols),
               names_to = "frame", values_to = "count") |>
  mutate(
    frame = str_remove(frame, "frame_"),
    pct   = count / n_total * 100
  ) |>
  filter(!is.na(year_month))

ggplot(frame_monthly, aes(x = year_month, y = pct, color = frame)) +
  geom_line(linewidth = 0.8, alpha = 0.8) +
  geom_smooth(method = "loess", se = FALSE, linewidth = 1.2, linetype = "dashed") +
  facet_wrap(~frame, ncol = 2, scales = "free_y") +
  scale_color_manual(values = frame_colors, guide = "none") +
  scale_x_date(date_breaks = "6 months", date_labels = "%b\n%Y") +
  labs(title = "Frame evolution",
       subtitle = "Monthly share of articles per frame with LOESS trend",
       x = NULL, y = "% articles") +
  theme(axis.text.x = element_text(size = 8))

Figure 10: Frame evolution over time

4.4 Frame co-occurrence

# Build frame matrix
frame_matrix <- as.matrix(corpus_data[, frame_cols])
colnames(frame_matrix) <- str_remove(colnames(frame_matrix), "frame_")

# Co-occurrence as proportion: of articles with frame A, what share also has frame B
n_articles <- nrow(frame_matrix)
n_frames <- ncol(frame_matrix)
cooc_matrix <- matrix(0, n_frames, n_frames,
                      dimnames = list(colnames(frame_matrix), colnames(frame_matrix)))

for (i in seq_len(n_frames)) {
  for (j in seq_len(n_frames)) {
    if (sum(frame_matrix[, i]) > 0) {
      cooc_matrix[i, j] <- sum(frame_matrix[, i] & frame_matrix[, j]) /
                            sum(frame_matrix[, i]) * 100
    }
  }
}

cooc_long <- as.data.frame(as.table(cooc_matrix)) |>
  setNames(c("Frame_A", "Frame_B", "pct"))

ggplot(cooc_long, aes(x = Frame_B, y = Frame_A, fill = pct)) +
  geom_tile(color = "white") +
  geom_text(aes(label = round(pct, 0)), size = 3) +
  scale_fill_gradient(low = "white", high = "#2c7bb6", name = "% co-occur") +
  labs(title = "Frame co-occurrence",
       subtitle = "Of articles containing row frame, what % also contain column frame",
       x = NULL, y = NULL) +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

Figure 11: Frame co-occurrence heatmap

4.5 Frame density per article

corpus_data$n_frames <- rowSums(corpus_data[, frame_cols], na.rm = TRUE)

frame_density <- corpus_data |>
  count(n_frames) |>
  mutate(pct = round(n / sum(n) * 100, 1))

ggplot(frame_density, aes(x = factor(n_frames), y = n)) +
  geom_col(fill = "#2c7bb6", alpha = 0.8) +
  geom_text(aes(label = paste0(pct, "%")), vjust = -0.5, size = 3.5) +
  labs(title = "Frame density per article",
       subtitle = "How many distinct frames appear within a single article",
       x = "Number of frames detected", y = "Articles") +
  expand_limits(y = max(frame_density$n) * 1.1)

Figure 12: Number of frames detected per article

4.6 Composite threat vs opportunity index

corpus_data$threat <- as.integer(
  corpus_data$frame_JOB_LOSS | corpus_data$frame_FEAR_RESISTANCE |
  corpus_data$frame_INEQUALITY
)
corpus_data$opportunity <- as.integer(
  corpus_data$frame_JOB_CREATION | corpus_data$frame_PRODUCTIVITY |
  corpus_data$frame_TRANSFORMATION
)

composite_monthly <- corpus_data |>
  group_by(year_month) |>
  summarise(
    n = n(),
    threat_pct = sum(threat, na.rm = TRUE) / n() * 100,
    opportunity_pct = sum(opportunity, na.rm = TRUE) / n() * 100,
    .groups = "drop"
  ) |>
  filter(!is.na(year_month)) |>
  pivot_longer(cols = c(threat_pct, opportunity_pct),
               names_to = "index", values_to = "pct") |>
  mutate(index = ifelse(str_detect(index, "threat"), "Threat", "Opportunity"))

ggplot(composite_monthly, aes(x = year_month, y = pct, color = index)) +
  geom_line(linewidth = 0.6, alpha = 0.5) +
  geom_smooth(method = "loess", span = 0.3, se = TRUE, linewidth = 1.2) +
  scale_color_manual(values = c("Threat" = "#e41a1c", "Opportunity" = "#4daf4a")) +
  labs(title = "Threat vs opportunity narrative",
       subtitle = "Threat = job loss + fear + inequality; Opportunity = creation + productivity + transformation",
       x = NULL, y = "% of articles", color = NULL)

Figure 13: Composite threat vs opportunity frame indices over time

5 Actor analysis

actor_dictionaries <- CONFIG$actors

for (actor_name in names(actor_dictionaries)) {
  pattern <- paste(actor_dictionaries[[actor_name]], collapse = "|")
  corpus_data[[paste0("actor_", actor_name)]] <- stri_detect_regex(
    corpus_data$.text_lower, pattern
  )
}

actor_cols <- paste0("actor_", names(actor_dictionaries))
actor_counts <- corpus_data |>
  summarise(across(all_of(actor_cols), sum, na.rm = TRUE)) |>
  pivot_longer(everything(), names_to = "actor", values_to = "count") |>
  mutate(
    actor = str_remove(actor, "actor_"),
    pct   = round(count / nrow(corpus_data) * 100, 1)
  ) |>
  arrange(desc(count))

kable(actor_counts, col.names = c("Actor", "Articles", "% corpus")) |>
  kable_styling(bootstrap_options = c("striped", "hover"), full_width = FALSE)

Actor	Articles	% corpus
EMPLOYERS	17570	52.1
TECH_COMPANIES	14652	43.5
EXPERTS	12873	38.2
WORKERS	11739	34.8
POLICY_MAKERS	9544	28.3
UNIONS	695	2.1

ggplot(actor_counts, aes(x = reorder(actor, count), y = count)) +
  geom_col(fill = "#377eb8", alpha = 0.8) +
  geom_text(aes(label = paste0(pct, "%")), hjust = -0.1, size = 3.5) +
  coord_flip() +
  labs(title = "Actor prevalence", x = NULL, y = "Articles") +
  expand_limits(x = max(actor_counts$count) * 1.15)

Figure 14: Actor prevalence

5.1 Actor evolution over time

actor_monthly <- corpus_data |>
  group_by(year_month) |>
  summarise(
    n_total = n(),
    across(all_of(actor_cols), sum, na.rm = TRUE),
    .groups = "drop"
  ) |>
  pivot_longer(cols = all_of(actor_cols),
               names_to = "actor", values_to = "count") |>
  mutate(
    actor = str_remove(actor, "actor_"),
    pct   = count / n_total * 100
  ) |>
  filter(!is.na(year_month))

ggplot(actor_monthly, aes(x = year_month, y = pct, color = actor)) +
  geom_line(linewidth = 0.7, alpha = 0.7) +
  geom_smooth(method = "loess", se = FALSE, linewidth = 1, linetype = "dashed") +
  facet_wrap(~ actor, ncol = 2, scales = "free_y") +
  scale_x_date(date_breaks = "6 months", date_labels = "%b\n%Y") +
  labs(title = "Actor prevalence over time",
       subtitle = "Monthly share with LOESS trend",
       x = NULL, y = "% of articles", color = NULL) +
  theme(axis.text.x = element_text(size = 8), legend.position = "none")

Figure 15: Actor prevalence over time

5.2 Actor-frame associations

actor_frame_assoc <- expand.grid(
  actor = names(actor_dictionaries),
  frame = names(frame_dictionaries),
  stringsAsFactors = FALSE
)

actor_frame_assoc$pct <- mapply(function(a, f) {
  a_col <- paste0("actor_", a)
  f_col <- paste0("frame_", f)
  actor_articles <- corpus_data[[a_col]]
  n_actor <- sum(actor_articles, na.rm = TRUE)
  if (n_actor == 0) return(0)
  sum(actor_articles & corpus_data[[f_col]], na.rm = TRUE) / n_actor * 100
}, actor_frame_assoc$actor, actor_frame_assoc$frame)

ggplot(actor_frame_assoc, aes(x = frame, y = actor, fill = pct)) +
  geom_tile(color = "white") +
  geom_text(aes(label = round(pct, 1)), size = 3) +
  scale_fill_gradient(low = "white", high = "#2c7bb6", name = "% of actor articles") +
  labs(title = "Actor-frame associations",
       subtitle = "Of articles mentioning the actor, what % also contain the frame",
       x = NULL, y = NULL) +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

Figure 16: Actor-frame association heatmap

6 Outlet classification

outlet_cfg <- CONFIG$outlet_types

corpus_data$outlet_type <- "Other"
for (type_name in names(outlet_cfg)) {
  for (pat in outlet_cfg[[type_name]]) {
    matches <- stri_detect_regex(stri_trans_tolower(corpus_data$FROM), pat)
    corpus_data$outlet_type[matches] <- type_name
  }
}

outlet_dist <- corpus_data |>
  count(outlet_type, sort = TRUE) |>
  mutate(pct = round(n / sum(n) * 100, 1))

kable(outlet_dist, col.names = c("Outlet type", "N", "%")) |>
  kable_styling(bootstrap_options = c("striped", "hover"), full_width = FALSE)

Outlet type	N	%
Other	22472	66.7
Quality	4702	14.0
Business	2436	7.2
Tabloid	1511	4.5
Tech	1235	3.7
Regional	1025	3.0
Public	311	0.9

min_articles <- CONFIG$analysis$min_articles_per_outlet

frames_by_outlet <- corpus_data |>
  group_by(outlet_type) |>
  summarise(
    n = n(),
    across(all_of(frame_cols), ~sum(.x, na.rm = TRUE) / n() * 100)
  ) |>
  filter(n >= min_articles) |>
  pivot_longer(cols = all_of(frame_cols),
               names_to = "frame", values_to = "pct") |>
  mutate(frame = str_remove(frame, "frame_"))

ggplot(frames_by_outlet, aes(x = frame, y = pct, fill = outlet_type)) +
  geom_col(position = "dodge", alpha = 0.8) +
  scale_fill_brewer(palette = "Set2") +
  labs(title = "Frame prevalence by outlet type",
       x = NULL, y = "% articles", fill = "Outlet type") +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

Figure 17: Frame prevalence by outlet type

6.1 Outlet volume over time

outlet_monthly <- corpus_data |>
  filter(outlet_type != "Other") |>
  count(year_month, outlet_type) |>
  filter(!is.na(year_month))

ggplot(outlet_monthly, aes(x = year_month, y = n, fill = outlet_type)) +
  geom_area(alpha = 0.7, position = "stack") +
  scale_fill_brewer(palette = "Set2") +
  scale_x_date(date_breaks = "3 months", date_labels = "%b\n%Y") +
  labs(title = "Coverage volume by outlet type",
       x = NULL, y = "Articles", fill = "Outlet type")

Figure 18: Monthly volume by outlet type

6.2 Threat/opportunity by outlet type

outlet_composite <- corpus_data |>
  filter(outlet_type != "Other") |>
  group_by(outlet_type) |>
  summarise(
    n = n(),
    threat_pct = sum(threat, na.rm = TRUE) / n() * 100,
    opportunity_pct = sum(opportunity, na.rm = TRUE) / n() * 100,
    .groups = "drop"
  ) |>
  mutate(ratio = round(threat_pct / pmax(opportunity_pct, 0.1), 2)) |>
  pivot_longer(cols = c(threat_pct, opportunity_pct),
               names_to = "index", values_to = "pct") |>
  mutate(index = ifelse(str_detect(index, "threat"), "Threat", "Opportunity"))

ggplot(outlet_composite, aes(x = outlet_type, y = pct, fill = index)) +
  geom_col(position = "dodge", alpha = 0.85) +
  scale_fill_manual(values = c("Threat" = "#e41a1c", "Opportunity" = "#4daf4a")) +
  labs(title = "Threat vs opportunity by outlet type",
       x = NULL, y = "% of articles", fill = NULL) +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

Figure 19: Threat vs opportunity framing by outlet type

7 Engagement analysis

if (all(c("REACH", "INTERACTIONS") %in% names(corpus_data))) {

  p_reach <- corpus_data |>
    filter(!is.na(REACH), REACH > 0,
           REACH < quantile(REACH, 0.99, na.rm = TRUE)) |>
    ggplot(aes(x = REACH)) +
    geom_histogram(bins = 50, fill = "#2c7bb6", alpha = 0.7, color = "white") +
    scale_x_log10(labels = scales::comma) +
    labs(title = "Article reach (log scale)", x = "Reach", y = "Articles")

  p_interactions <- corpus_data |>
    filter(!is.na(INTERACTIONS), INTERACTIONS > 0,
           INTERACTIONS < quantile(INTERACTIONS, 0.99, na.rm = TRUE)) |>
    ggplot(aes(x = INTERACTIONS)) +
    geom_histogram(bins = 50, fill = "#e41a1c", alpha = 0.7, color = "white") +
    scale_x_log10(labels = scales::comma) +
    labs(title = "Article interactions (log scale)", x = "Interactions", y = "Articles")

  p_reach | p_interactions
}

Figure 20: Distribution of article reach and interactions

if (all(c("REACH", "INTERACTIONS") %in% names(corpus_data))) {

  engagement_by_frame <- lapply(names(frame_dictionaries), function(fname) {
    f_col <- paste0("frame_", fname)
    articles_with <- corpus_data |> filter(.data[[f_col]] == TRUE)

    tibble(
      Frame = fname,
      N = nrow(articles_with),
      Median_reach = round(median(articles_with$REACH, na.rm = TRUE)),
      Median_interactions = round(median(articles_with$INTERACTIONS, na.rm = TRUE)),
      Mean_reach = round(mean(articles_with$REACH, na.rm = TRUE)),
      Mean_interactions = round(mean(articles_with$INTERACTIONS, na.rm = TRUE))
    )
  })

  engagement_tbl <- bind_rows(engagement_by_frame) |> arrange(desc(Median_reach))

  kable(engagement_tbl,
        col.names = c("Frame", "N", "Median Reach", "Median Interactions",
                       "Mean Reach", "Mean Interactions")) |>
    kable_styling(bootstrap_options = c("striped", "hover"), full_width = FALSE)
}

Table 2: Median reach and interactions by dominant frame

Frame	N	Median Reach	Median Interactions	Mean Reach	Mean Interactions
INEQUALITY	664	533	2	2202	59
FEAR_RESISTANCE	5695	480	2	2631	72
JOB_LOSS	651	450	2	1883	40
SKILLS	2372	450	2	3160	98
JOB_CREATION	1550	435	1	3031	87
REGULATION	1519	434	1	2622	57
TRANSFORMATION	1755	280	0	1559	41
PRODUCTIVITY	7748	261	1	1997	53

if (all(c("REACH", "INTERACTIONS") %in% names(corpus_data))) {

  engagement_platform <- corpus_data |>
    filter(platform %in% platforms_with_data) |>
    group_by(platform) |>
    summarise(
      n = n(),
      median_reach = round(median(REACH, na.rm = TRUE)),
      median_interactions = round(median(INTERACTIONS, na.rm = TRUE)),
      total_reach = round(sum(REACH, na.rm = TRUE)),
      total_interactions = round(sum(INTERACTIONS, na.rm = TRUE)),
      .groups = "drop"
    ) |>
    arrange(desc(median_reach))

  kable(engagement_platform,
        col.names = c("Platform", "N", "Median Reach", "Median Interactions",
                       "Total Reach", "Total Interactions")) |>
    kable_styling(bootstrap_options = c("striped", "hover"), full_width = FALSE)
}

Table 3: Engagement metrics by platform

Platform	N	Median Reach	Median Interactions	Total Reach	Total Interactions
Facebook	727	1243	8	6223726	43419
web	31927	360	1	68919863	1742228
YouTube	433	99	3	1115866	14129
Twitter	67	86	0	25662	320
Reddit	256	NA	NA	0	0
forum	269	NA	NA	0	0

7.1 Top articles by reach

if ("REACH" %in% names(corpus_data)) {
  top_reach <- corpus_data |>
    filter(!is.na(REACH)) |>
    arrange(desc(REACH)) |>
    head(20) |>
    dplyr::select(DATE, TITLE, FROM, REACH, INTERACTIONS) |>
    mutate(
      TITLE = substr(TITLE, 1, 80),
      REACH = format(REACH, big.mark = ","),
      INTERACTIONS = format(INTERACTIONS, big.mark = ",")
    )

  kable(top_reach, col.names = c("Date", "Title", "Source", "Reach", "Interactions")) |>
    kable_styling(bootstrap_options = c("striped", "hover", "condensed"),
                  full_width = FALSE, font_size = 11) |>
    scroll_box(height = "400px")
}

Table 4: Top 20 articles by estimated reach

Date	Title	Source	Reach	Interactions
2021-03-12	Uključi se u e-Savjetovanja!	gov.hr	1,606,086	73,008
2023-09-07	DIGITAL HUMAN tema je jedne od najznačajnijih međunarodnih znanstvenih konferenc	mnovine.hr	1,280,448	29,222
2023-10-30	Lovci i forenzičari, crveni i plavi timovi	mnovine.hr	1,280,448	29,222
2023-11-21	Međimurske novine	mnovine.hr	1,280,448	29,222
2022-07-26	Kakav algoritam koriste online casino igre u Hrvatskoj?	dnevno.hr	1,133,925	37,005
2021-11-09	TRAŽITE NOVI POSAO I NOVE IZAZOVE? Možda ovi poslodavci trebaju baš vas, a neki	dnevno.hr	1,132,440	36,944
2023-02-01	Predstavljamo novi	bing.com	863,439	81
2023-02-07	Predstavljamo novi	bing.com	863,439	81
2023-06-05	[PODIJELI] Oda lažnim novinarima, Sanji Modrić, Renati Rašović, Borisu Trupčević	Građani za Mislava Kolakušića	818,766	13,373
2023-10-23	U posljednjih godinu dana termine poput "AI-a" i "ChatGPT-a" gotovo je nemoguće	Mastercard	789,464	362
2022-10-03	SEO OPTIMIZACIJA	tolo-design.com	780,147	29,439
2023-10-02	Umjetna inteligencija i budućnost automobilske industrije? Evo što trebate znati	TotalEnergies	562,239	1,115
2022-09-03	Kupi moju kuću - Službena Netflixova stranica	netflix.com	542,100	0
2022-04-14	ASUS Hrvatska - ExpertBook B1 (B1500) - Recenzija	ASUS Hrvatska	364,263	7
2021-05-25	Menadžment Turističkog Smještaja	direct-booker.com	353,265	12,681
2023-09-12	Croatia Records	crorec.net	316,770	7,104
2021-09-29	‘Plaće i penzije rastu, vraćaju nam se Hrvati iz inozemstva, doseljavaju stranci	jutarnji.hr	266,454	14,076
2021-08-26	Vlasnik firme u Zagrebu uvodi radni tjedan od četiri dana, plaće ostaju iste	index.hr	266,379	22,243
2023-06-24	‎„GlowCast - osobni profesionalni brending kao jedina stabilna valuta u svijetu	|apple.com	|264,300	|0
2022-07-01	Uključi se u e-Savjetovanja!	gov.hr	217,917	11,315

8 Sentiment analysis

if ("AUTO_SENTIMENT" %in% names(corpus_data)) {
  sentiment_dist <- corpus_data |>
    count(AUTO_SENTIMENT, sort = TRUE) |>
    mutate(pct = round(n / sum(n) * 100, 1))

  kable(sentiment_dist, col.names = c("Sentiment", "N", "%")) |>
    kable_styling(bootstrap_options = c("striped", "hover"), full_width = FALSE)
} else {
  cat("AUTO_SENTIMENT column not available in corpus.\n")
}

Sentiment	N	%
positive	22644	67.2
negative	5973	17.7
neutral	5075	15.1

if ("AUTO_SENTIMENT" %in% names(corpus_data)) {
  sentiment_monthly <- corpus_data |>
    filter(!is.na(AUTO_SENTIMENT) & !is.na(year_month)) |>
    count(year_month, AUTO_SENTIMENT) |>
    group_by(year_month) |>
    mutate(pct = n / sum(n)) |>
    ungroup()

  ggplot(sentiment_monthly, aes(x = year_month, y = pct, fill = AUTO_SENTIMENT)) +
    geom_area(alpha = 0.7) +
    scale_fill_manual(values = sentiment_colors) +
    scale_y_continuous(labels = scales::percent) +
    scale_x_date(date_breaks = "3 months", date_labels = "%b\n%Y") +
    labs(title = "Sentiment over time",
         x = NULL, y = "Share", fill = "Sentiment")
}

Figure 21: Sentiment distribution over time

8.1 Sentiment by frame

if ("AUTO_SENTIMENT" %in% names(corpus_data)) {

  sentiment_frame <- lapply(names(frame_dictionaries), function(fname) {
    f_col <- paste0("frame_", fname)
    corpus_data |>
      filter(.data[[f_col]] == TRUE, !is.na(AUTO_SENTIMENT)) |>
      count(AUTO_SENTIMENT) |>
      mutate(
        frame = fname,
        pct = n / sum(n) * 100
      )
  })

  sentiment_frame_df <- bind_rows(sentiment_frame)

  ggplot(sentiment_frame_df,
         aes(x = frame, y = pct, fill = AUTO_SENTIMENT)) +
    geom_col(alpha = 0.85) +
    scale_fill_manual(values = sentiment_colors) +
    labs(title = "Sentiment composition by frame",
         subtitle = "Some frames carry systematically more negative sentiment",
         x = NULL, y = "% of articles", fill = "Sentiment") +
    theme(axis.text.x = element_text(angle = 45, hjust = 1))
}

Figure 22: Sentiment composition by frame

8.2 Sentiment by outlet type

if ("AUTO_SENTIMENT" %in% names(corpus_data)) {

  sentiment_outlet <- corpus_data |>
    filter(!is.na(AUTO_SENTIMENT), outlet_type != "Other") |>
    count(outlet_type, AUTO_SENTIMENT) |>
    group_by(outlet_type) |>
    mutate(pct = n / sum(n) * 100) |>
    ungroup()

  ggplot(sentiment_outlet, aes(x = outlet_type, y = pct, fill = AUTO_SENTIMENT)) +
    geom_col(alpha = 0.85) +
    scale_fill_manual(values = sentiment_colors) +
    labs(title = "Sentiment by outlet type",
         x = NULL, y = "% of articles", fill = "Sentiment") +
    theme(axis.text.x = element_text(angle = 45, hjust = 1))
}

Figure 23: Sentiment distribution by outlet type

9 Keyword analysis

9.1 Most frequent AI terms

ai_keywords <- list(
  "umjetna inteligencija" = "umjetn.*inteligencij",
  "strojno učenje"        = "strojn.*učenj",
  "ChatGPT"               = "chat.?gpt",
  "GPT-4/GPT-3"           = "gpt.?[34]",
  "OpenAI"                = "openai|open ai",
  "generativni AI"        = "generativn.*(ai|umjetn)",
  "automatizacija"        = "automatizacij",
  "robotizacija"          = "robotizacij",
  "algoritam"             = "algoritm",
  "neuronska mreža"       = "neuronsk|neuralna",
  "chatbot"               = "chatbot",
  "LLM"                   = "\\bllm\\b",
  "Gemini"                = "\\bgemini\\b",
  "Copilot"               = "copilot",
  "duboko učenje"         = "duboko.*učenj"
)

ai_freq <- sapply(ai_keywords, function(pat) {
  sum(stri_detect_regex(corpus_data$.text_lower, pat), na.rm = TRUE)
})

ai_freq_tbl <- tibble(
  Term = names(ai_freq),
  Articles = unname(ai_freq),
  Pct = round(ai_freq / nrow(corpus_data) * 100, 1)
) |> arrange(desc(Articles))

kable(ai_freq_tbl, col.names = c("AI Term", "Articles", "% corpus")) |>
  kable_styling(bootstrap_options = c("striped", "hover"), full_width = FALSE)

Table 5: Frequency of individual AI keyword matches

AI Term	Articles	% corpus
umjetna inteligencija	22637	67.2
automatizacija	7521	22.3
ChatGPT	3150	9.3
strojno učenje	3095	9.2
algoritam	2221	6.6
chatbot	2042	6.1
OpenAI	1661	4.9
generativni AI	1612	4.8
robotizacija	1010	3.0
neuronska mreža	893	2.7
GPT-4/GPT-3	508	1.5
duboko učenje	467	1.4
LLM	218	0.6
Copilot	183	0.5
Gemini	44	0.1

9.2 Most frequent labour terms

labour_keywords <- list(
  "posao/poslovi"           = "\\bposao|\\bposlovi|\\bposlove|\\bposlova",
  "zaposleni/zapošljavanje" = "zaposlen|zapošljav",
  "nezaposlenost"           = "nezaposlen",
  "radno mjesto"            = "radn.*mjest",
  "tržište rada"            = "tržišt.*rada",
  "vještine"                = "vještin",
  "kompetencije"            = "kompetencij",
  "karijera"                = "karijer",
  "produktivnost"           = "produktivnost",
  "otpuštanje"              = "otpuštan",
  "prekvalifikacija"        = "prekvalifikacij|dokvalifikacij",
  "plaća"                   = "\\bplaća|\\bplaće|\\bplaću",
  "poslodavac"              = "poslodav",
  "radna snaga"             = "radn.*snag",
  "zanimanje"               = "zaniman"
)

labour_freq <- sapply(labour_keywords, function(pat) {
  sum(stri_detect_regex(corpus_data$.text_lower, pat), na.rm = TRUE)
})

labour_freq_tbl <- tibble(
  Term = names(labour_freq),
  Articles = unname(labour_freq),
  Pct = round(labour_freq / nrow(corpus_data) * 100, 1)
) |> arrange(desc(Articles))

kable(labour_freq_tbl, col.names = c("Labour Term", "Articles", "% corpus")) |>
  kable_styling(bootstrap_options = c("striped", "hover"), full_width = FALSE)

Table 6: Frequency of individual labour market keyword matches

Labour Term	Articles	% corpus
posao/poslovi	21881	64.9
zaposleni/zapošljavanje	12188	36.2
radno mjesto	8080	24.0
vještine	7078	21.0
plaća	6050	18.0
tržište rada	4348	12.9
karijera	4218	12.5
radna snaga	3275	9.7
produktivnost	2949	8.8
poslodavac	2528	7.5
zanimanje	2217	6.6
kompetencije	1955	5.8
nezaposlenost	987	2.9
otpuštanje	580	1.7
prekvalifikacija	270	0.8

9.3 AI term evolution

key_ai_terms <- list(
  "ChatGPT"      = "chat.?gpt",
  "AI/UI"        = "umjetn.*inteligencij",
  "automatizacija" = "automatizacij",
  "algoritam"    = "algoritm",
  "robotizacija" = "robotizacij"
)

ai_term_monthly <- lapply(names(key_ai_terms), function(term_name) {
  corpus_data |>
    group_by(year_month) |>
    summarise(
      pct = sum(stri_detect_regex(.text_lower, key_ai_terms[[term_name]]),
                na.rm = TRUE) / n() * 100,
      .groups = "drop"
    ) |>
    filter(!is.na(year_month)) |>
    mutate(term = term_name)
})

ai_term_monthly_df <- bind_rows(ai_term_monthly)

ggplot(ai_term_monthly_df, aes(x = year_month, y = pct, color = term)) +
  geom_line(linewidth = 0.7) +
  labs(title = "AI term prevalence over time",
       subtitle = "Share of articles mentioning each term",
       x = NULL, y = "% of articles", color = "Term") +
  scale_x_date(date_breaks = "3 months", date_labels = "%b\n%Y")

Figure 24: Selected AI term mention rates over time

10 Frame dynamics by platform

platform_frames <- corpus_data |>
  filter(platform %in% platforms_with_data) |>
  group_by(platform) |>
  summarise(
    n = n(),
    across(all_of(frame_cols), ~ sum(.x, na.rm = TRUE) / n() * 100),
    .groups = "drop"
  ) |>
  pivot_longer(cols = all_of(frame_cols),
               names_to = "frame", values_to = "pct") |>
  mutate(frame = str_remove(frame, "frame_"))

ggplot(platform_frames, aes(x = frame, y = pct, fill = platform)) +
  geom_col(position = "dodge", alpha = 0.85) +
  scale_fill_manual(values = platform_colors) +
  labs(title = "Frame prevalence by platform",
       x = NULL, y = "% of articles", fill = "Platform") +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

Figure 25: Frame prevalence by platform

ggplot(platform_frames, aes(x = frame, y = platform, fill = pct)) +
  geom_tile(color = "white") +
  geom_text(aes(label = round(pct, 1)), size = 3) +
  scale_fill_gradient2(low = "white", mid = "#abd9e9", high = "#2c7bb6",
                       midpoint = median(platform_frames$pct)) +
  labs(title = "Platform x frame heatmap",
       x = NULL, y = NULL, fill = "% of articles") +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

Figure 26: Platform x frame heatmap

11 Sample articles

set.seed(CONFIG$analysis$seed)
sample_articles <- corpus_data |>
  slice_sample(n = min(30, nrow(corpus_data))) |>
  dplyr::select(DATE, TITLE, FROM) |>
  arrange(DATE)

kable(sample_articles, col.names = c("Date", "Title", "Source")) |>
  kable_styling(bootstrap_options = c("striped", "hover", "condensed"),
                full_width = FALSE, font_size = 11) |>
  scroll_box(height = "400px")

Date	Title	Source
2021-01-03	Produktivnost bez BS-a	split-techcity.com
2021-01-07	Serengeti želi rast u zapadnoj Europi, a 2020. završavaju na sličnim razinama prihoda i dobiti kao 2019.	ictbusiness.info
2021-01-10	Stručnjaci predviđaju da bi u idućih 20 godina ovih 12 poslova moglo nestati. Je li i vaš među njima?	sbonline.net
2021-08-20	Musk predstavio humanoidnog robota: ‘Rad će u budućnosti biti stvar izbora pojedinca‘	jutarnji.hr
2021-09-23	Huawei predstavio nadolazeće trendove u idućem desetljeću	seebiz.eu
2021-11-14	Aktualni trendovi u digitalizaciji Hrvatske, Unije - DESI	grabancijas.com
2021-11-24	Novi poslovi u Osijeku i okolici: Traže Vas Thermia kamini, Ancona grupa, Mehanotehna...	osijekexpress.com
2022-01-11	'Sve na 15 minuta', inteligentna mobilnost... Što Zagreb može prigrliti već za 8 godina? - Poslovni dnevnik	poslovni.hr
2022-01-24	Tko je nasljednica Dubravke Vrgoč? Opernu primadonu već slave u HNK, a u velikom intervju otkrila je ambiciozne planove	zagreb.info
2022-01-27	AC Group: Vlastito iskustvo i znanje pretočili u razvoj uspješne obiteljske firme	varazdinske-vijesti.hr
2022-02-22	ManpowerGroup SEE uvodi četverodnevni radni tjedan	jatrgovac.com
2022-03-21	poslovniFM	poslovnifm.com
2022-04-27	Samsung Galaxy S22 serija pametnih telefona i dobar AI su dokaz besprijekorne kvalitete	hcl.hr
2022-06-09	RITTAL i KONČAR - Digital partneri u izgradnji podatkovnih centara i digitalizaciji regije - Poslovni dnevnik	poslovni.hr
2022-08-08	Kurkumin - polifenol koji ne prestaje biti zagonetka	inpharma.hr
2022-09-16	Matej Dujmović: Tvrtke moraju početi razmišljati o automatizaciji skladišnih i proizvodnih procesa	lidermedia.hr
2022-11-24	Preko 70 stručnih predavača 1.12. u Zagrebu na konferenciji koja mijenja sve!	24sata.hr
2023-01-06	Terra Meera - čudesno mjesto mira i regeneracije u dalmatinskom zaleđu	grazia.hr
2023-04-06	Bowery Farming - integrirane farme	gospodarski.hr
2023-04-29	Ključno je postići suživot ljudi i strojeva, odnosno umjetne inteligencije	itnovosti.com
2023-05-27	Sektori industrije i infrastrukture s najboljim izgledima	seebiz.eu
2023-06-08	Snowden ponovno upozorava svijet! Njegovo otkriće je 'dječja igra' u odnosu na to što danas rade ljudima: Nadzor otišao predaleko	dnevno.hr
2023-06-30	sudjelovalo na predstavljanju Europskog digitalnog centra inovacija CROBOHUB++ - Sveučilišni računski centar (Srce)	unizg.hr
2023-07-17	Vijeće sigurnosti UN-a održat će prvu sjednicu o umjetnoj inteligenciji	novilist.hr
2023-07-25	Appleova AI mogla bi značajno poboljšati iPhone, evo i kako	dnevnik.hr
2023-07-26	Tomislav Miletić - Trgovci Novim I Rabljim Automobilima Doživjet Će Darqinovu Evoluciju	lidermedia.hr
2023-08-30	PC Chip	pcchip.hr
2023-09-28	OpenAI bi mogao vrijediti gotovo 100 milijardi dolara	tportal.hr
2023-10-20	M. Kepec: Kako postati uspješan DevOps inženjer	glas-slavonije.hr
2023-11-20	Prošli tjedan smo kroz naš PPK (PAR Poduzetnički Kamp) uronili u svijet poduzetništva! 🚀 Započeli smo s predavanjem o temi, formirali timove i započeli... |	eleučilište PAR |

12 Summary

n_articles    <- nrow(corpus_data)
date_range    <- paste(min(corpus_data$DATE), "to", max(corpus_data$DATE))
n_sources     <- n_distinct(corpus_data$FROM)
top_frame     <- frame_counts$frame[1]
top_frame_pct <- frame_counts$pct[1]

# Frame summary
cat("ANALYSIS SUMMARY\n")

ANALYSIS SUMMARY

cat("================\n\n")

================

cat("Total articles:", format(n_articles, big.mark = ","), "\n")

Total articles: 33,692 

cat("Period:", date_range, "\n")

Period: 2021-01-01 to 2023-12-31 

cat("Sources:", format(n_sources, big.mark = ","), "\n")

Sources: 2,034 

cat("Dominant frame:", top_frame, "(", top_frame_pct, "% of articles)\n\n")

Dominant frame: PRODUCTIVITY ( 23 % of articles)

# Frame density
cat("Frame density:\n")

Frame density:

cat("  Articles with 0 frames:", sum(corpus_data$n_frames == 0), "\n")

  Articles with 0 frames: 17781 

cat("  Articles with 1 frame:", sum(corpus_data$n_frames == 1), "\n")

  Articles with 1 frame: 11215 

cat("  Articles with 2+ frames:", sum(corpus_data$n_frames >= 2), "\n\n")

  Articles with 2+ frames: 4696 

# Threat vs opportunity
cat("Composite indices:\n")

Composite indices:

cat("  Articles with threat framing:", sum(corpus_data$threat, na.rm = TRUE),
    "(", round(mean(corpus_data$threat, na.rm = TRUE) * 100, 1), "%)\n")

  Articles with threat framing: 6433 ( 19.1 %)

cat("  Articles with opportunity framing:", sum(corpus_data$opportunity, na.rm = TRUE),
    "(", round(mean(corpus_data$opportunity, na.rm = TRUE) * 100, 1), "%)\n")

  Articles with opportunity framing: 9685 ( 28.7 %)

summary_comprehensive <- tibble(
  Category = c(
    rep("Corpus", 5),
    rep("Frames", 4),
    rep("Actors", 2),
    rep("Platforms", 2)
  ),
  Metric = c(
    "Total articles", "Unique sources", "Date range",
    "Median article length (words)", "Mean article length (words)",
    "Dominant frame", "Articles with any frame",
    "Mean frames per article", "Threat/opportunity ratio",
    "Most mentioned actor", "Least mentioned actor",
    "Dominant platform", "Number of platforms"
  ),
  Value = c(
    format(nrow(corpus_data), big.mark = ","),
    format(n_distinct(corpus_data$FROM), big.mark = ","),
    paste(min(corpus_data$DATE), "to", max(corpus_data$DATE)),
    round(median(corpus_data$word_count, na.rm = TRUE)),
    round(mean(corpus_data$word_count, na.rm = TRUE)),
    paste0(frame_counts$frame[1], " (", frame_counts$pct[1], "%)"),
    paste0(sum(corpus_data$n_frames > 0), " (",
           round(mean(corpus_data$n_frames > 0) * 100, 1), "%)"),
    round(mean(corpus_data$n_frames), 2),
    round(sum(corpus_data$threat, na.rm = TRUE) /
          max(sum(corpus_data$opportunity, na.rm = TRUE), 1), 2),
    paste0(actor_counts$actor[1], " (", actor_counts$pct[1], "%)"),
    paste0(actor_counts$actor[nrow(actor_counts)], " (",
           actor_counts$pct[nrow(actor_counts)], "%)"),
    paste0(source_dist$SOURCE_TYPE[1], " (", source_dist$pct[1], "%)"),
    length(platforms_with_data)
  )
)

kable(summary_comprehensive, col.names = c("Category", "Metric", "Value")) |>
  kable_styling(bootstrap_options = c("striped", "hover"), full_width = FALSE) |>
  pack_rows(index = table(summary_comprehensive$Category)[
    unique(summary_comprehensive$Category)
  ])

Table 7: Comprehensive corpus statistics

Category	Metric	Value
Corpus
Corpus	Total articles	33,692
Corpus	Unique sources	2,034
Corpus	Date range	2021-01-01 to 2023-12-31
Corpus	Median article length (words)	646
Corpus	Mean article length (words)	899
Frames
Frames	Dominant frame	PRODUCTIVITY (23%)
Frames	Articles with any frame	15911 (47.2%)
Frames	Mean frames per article	0.65
Frames	Threat/opportunity ratio	0.66
Actors
Actors	Most mentioned actor	EMPLOYERS (52.1%)
Actors	Least mentioned actor	UNIONS (2.1%)
Platforms
Platforms	Dominant platform	web (94.8%)
Platforms	Number of platforms	6

13 Data export

export_data <- corpus_data |> dplyr::select(-`.text_lower`)
saveRDS(export_data, path_analysed_corpus)

14 Session info

sessionInfo()

R version 4.5.2 (2025-10-31 ucrt)
Platform: x86_64-w64-mingw32/x64
Running under: Windows 11 x64 (build 22631)

Matrix products: default
  LAPACK version 3.12.1

locale:
[1] LC_COLLATE=Croatian_Croatia.utf8  LC_CTYPE=Croatian_Croatia.utf8   
[3] LC_MONETARY=Croatian_Croatia.utf8 LC_NUMERIC=C                     
[5] LC_TIME=Croatian_Croatia.utf8    

time zone: Europe/Zagreb
tzcode source: internal

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
 [1] tidytext_0.4.3            quanteda.textstats_0.97.2
 [3] quanteda_4.3.1            kableExtra_1.4.0         
 [5] knitr_1.50                ggrepel_0.9.6            
 [7] patchwork_1.3.2           scales_1.4.0             
 [9] ggplot2_4.0.1             tibble_3.3.0             
[11] forcats_1.0.1             lubridate_1.9.4          
[13] stringi_1.8.7             stringr_1.6.0            
[15] tidyr_1.3.1               dplyr_1.1.4              
[17] yaml_2.3.11              

loaded via a namespace (and not attached):
 [1] janeaustenr_1.0.0  generics_0.1.4     xml2_1.5.1         lattice_0.22-7    
 [5] digest_0.6.39      magrittr_2.0.4     evaluate_1.0.5     grid_4.5.2        
 [9] timechange_0.3.0   RColorBrewer_1.1-3 fastmap_1.2.0      jsonlite_2.0.0    
[13] Matrix_1.7-4       mgcv_1.9-3         stopwords_2.3      purrr_1.2.0       
[17] viridisLite_0.4.2  textshaping_1.0.4  cli_3.6.5          rlang_1.1.6       
[21] tokenizers_0.3.0   splines_4.5.2      withr_3.0.2        tools_4.5.2       
[25] fastmatch_1.1-6    vctrs_0.6.5        R6_2.6.1           lifecycle_1.0.4   
[29] htmlwidgets_1.6.4  pkgconfig_2.0.3    pillar_1.11.1      gtable_0.3.6      
[33] glue_1.8.0         Rcpp_1.1.0         systemfonts_1.3.1  xfun_0.54         
[37] tidyselect_1.2.1   rstudioapi_0.17.1  dichromat_2.0-0.1  farver_2.1.2      
[41] nlme_3.1-168       SnowballC_0.7.1    htmltools_0.5.8.1  labeling_0.4.3    
[45] rmarkdown_2.30     svglite_2.2.2      compiler_4.5.2     nsyllable_1.0.1   
[49] S7_0.2.1